⭐ Livepeer Agent MCP · Long-form → social · Dubbing

83,000 brain scans, in three languages A 14-minute TEDx talk becomes a 90-second vertical highlight reel — then dubbed into Chinese, French, and Spanish, with translated captions and a music bed. One paste, one pass, per language.

Source: Dr. Daniel Amen — "The most important lesson from 83,000 brain scans" (TEDxOrangeCoast, ~14½ min). The agent found the key moments, cleaned the filler sounds, reframed to 9:16, wrote editorial key-message cards, burned captions, and laid a soft score — then cloned the speaker's voice and re-spoke the whole reel in each target language. This is the long-form → social playbook, run end to end.

source
TEDx · ~14½ min
output
~85s · 9:16 · ×3 languages
voice
FR/ES cloned · ZH matched
caps used
wizper · chatterbox-tts · music · ffmpeg
cost
~$0.16 all in

Three cuts — same talk, same key messages, three languages

🇨🇳 Chinese中文
Translated speech + captions · matched voice (Mandarin isn't clone-supported yet)
🇫🇷 Frenchfrançais
Translated speech + captions · the speaker's own cloned voice
🇪🇸 Spanishespañol
Translated speech + captions · the speaker's own cloned voice

Every version keeps the same 5-beat arc — psychiatry doesn't look → no two brains are alike → you can change a brain → the proof (a boy, a cyst) → you are not stuck — with the cards, captions, and now the spoken words all localized.

How it was made — the playbook, step by step

01
Find the moments
Transcribe the talk (wizper) and pick the 5 strongest key messages.
02
Clean it up
Cut the "um / uh / er" and dead air so each line is tight.
03
Reframe 9:16
Speaker centered on a blurred fill — nothing cropped off.
04
Translate
Each beat into Chinese, French, and Spanish.
05
Clone the voice
chatterbox-tts re-speaks each line in the speaker's own voice (FR/ES).
06
Caption + card
Translated burned captions + editorial key-message cards.
07
Score it
A generated music bed, mixed low under the voice.
08
Stitch
Cross-dissolve every beat into a single ~85s reel, per language.

Honest notes. This build swaps the audio (no lip-sync — the lips don't track the new language; fine for a caption-forward highlight reel). Chinese uses a high-quality matched Mandarin voice, not the cloned timbre — chatterbox-tts clones Latin-script voices (French/Spanish here) but falls back for Mandarin. Add veed-lipsync-v2 for lip-matched dubbing when you want it. Source is a public TEDx talk, credited on-screen — confirm rights before publishing.

Want this for your own long video? Fill in the long-form → social playbook — pick your languages, describe a music bed, and it hands you a ready-to-paste prompt.